This thesis is about the psychology of information selection and reasoning. It
investigates the way in which the probability of information influences the selection of information. Information which is expected to reduce uncertainty the most in a given probabilistic context is assumed to be the most relevant information to select.
Several computer-designed studies, two of which vary the probability of information in a learning phase, test the precise predictions and assumptions of the Oaksford and Chater (1994) model of optimal data selection, which specifically explains card selections in Wason's (1966) four card problem or selection task in terms of probability-dependent optimal data selection. This way of explaining reasoning in the selection task contrasts with traditional reasoning theories and explanations which assume that a reasoner's goal in the selection task is to falsify, and that truth-preserving rules of inferences, for example, logical deduction, underlie the inferential processing component of behaviour in this reasoning task.
In order to compare and contrast the O&C optimality approach with other theories in the psychology of reasoning, major theories of reasoning, as well as specific explanations of reasoning in the selection task, are reviewed. General optimality approaches to cognition are also reviewed (Stephens and Krebs, 1986) in order to place in proper theoretical context the O&C ( 1994) model of optimal data selection and how it explains, in particular, affirmative abstract versions of the selection task.
It is concluded that, at a psychological level, the O&C model of optimal data selection contributes significantly to the theoretical understanding of human reasoning because it proposes that a simple, adaptive, optimality-preserving decision rule (i.e. to select optimal data) governs the selection of information and what is perceived as optimal will change in different probabilistic contexts. Experimental results support an optimal data approach and demonstrate that it is not necessary for counterexamples to be represented in order to produce apparent "falsificationist" behaviour, as simply manipulating the probability of p and q in a learning phase prior to a selection phase can change, in accordance with precise predictions, selection task performance. At an optimality model level, because it adopts certain classical optimality assumptions regarding the optimality-preserving decision rule governing selection behaviour, the O&C model has similar strengths and weaknesses as simple optimality approaches,
and these are discussed.